This thesis examines online radicalisation as a multidimensional phenomenon shaped by psychological, social, and technological mechanisms and evaluates the legal frameworks designed to address the associated risks. The thesis examines the relationship between online radicalisation, disinformation, algorithmic recommendation systems, and platform governance, with particular emphasis on TikTok as a case study.

The theoretical part analyses key concepts including radicalisation, extremism, violent extremism, and disinformation. It explores how online environments may accelerate existing psychological and social dynamics through mechanisms such as the online disinhibition effect, echo chambers, filter bubbles, and the rabbit hole effect. Particular attention is devoted to the role of disinformation as both a by-product of algorithmic amplification and a deliberate instrument in coordinated hybrid operations.

The legal analysis examines the evolution of the European Union's approach to platform regulation, from voluntary self-regulatory initiatives to the binding obligations introduced by the Digital Services Act. The thesis evaluates the DSA framework with regard to systemic risk mitigation, transparency requirements, and researchers' access to platform data, while also discussing several practical challenges related to its implementation and enforcement.

The practical part consists of two complementary components. The first is the development of an OSINT toolkit implemented in Python using the Playwright browser automation framework. The toolkit supports automated metadata extraction and comment collection from TikTok profiles and individual videos, retrieving comment data by intercepting TikTok's internal API endpoint within the browser context. Collected data is stored in a structured JSON format designed to ensure consistency and reproducibility across experimental runs. The second component is a controlled observational case study examining recommendation dynamics using three newly created TikTok accounts exposed to distinct thematic categories related to political disinformation, pro-Russian propaganda, and extremist narratives.

The results show that TikTok's recommendation algorithm narrowed content feeds toward the trained thematic category within approximately 60 minutes of targeted interaction. Comment analysis revealed echo-chamber dynamics, with dissenting views receiving high reply volume but minimal approval, and cross-topic contamination was observed across thematic categories. The findings further suggest that algorithmic amplification can facilitate conditions associated with the early stages of online radicalisation, although radicalisation itself remains influenced by broader social and individual factors. The thesis concludes that while the DSA represents an important step towards greater platform accountability, significant challenges related to transparency, oversight, and algorithmic governance remain unresolved, particularly in the context of recommendation systems and algorithmic amplification.